Learning a discriminative feature for object detection based on feature fusing and context learning.

In this paper, we propose two discriminative feature learning strategies for object detection. The first involves a multi-layer features fusion strategy that improves the object detection performance mostly in terms of image size and aspect ratio. The second is context learning, which connects the multi-layer features with their location information. The experimental results on Pascal VOC 2007 and VOC 2012 show that our strategies improve the final detection results. In the future, we will continue to improve our object detection method in terms of both candidate region generation and feature representation learning.